Understanding and simplifying the structural similarity metric

The structural similarity (SSIM) metric and its multi-scale ex-tension (MS-SSIM) evaluate visual quality with a modied local measure of spatial correlation consisting of three com-ponents: mean, variance, and cross-correlation. This paper investigates how the SSIM components contribute to its qual-ity evaluation of common image artifacts. The predictive per-formance of the individual components and pairwise compo-nent products is assessed using the LIVE image database. Af-ter a nonlinear mapping, the product of the variance and cross-correlation components yields nearly identical linear correla-tion with subjective ratings as the complete SSIM and MS-SSIM computations. A computationally simple alternative to SSIM (c.f. Eq. (6)) that ignores the mean component and sets the local average patch values to 128 exhibits a 1 % decrease in linear correlation with subjective ratings to 0.934 from the complete SSIM evaluation with an over 20 % reduction in the number of multiplications. Index Terms — quality assessment, human visual system 1